Fine-Tuning Large Language Models With Declarative ML Orchestration
summary
Shows a reproducible pipeline that actually fine‑tunes a Llama 2 model with 8‑bit quantization and LoRA updates, declares per‑task CPU/GPU/memory on Kubernetes, and runs a DAG of tasks that train, quantize and save to Hugging Face. Uses Flyte for declarative orchestration so provisioning happens per step, demonstrates zero redundancy optimization and task‑level resource control, and the recorded demo runs end‑to‑end to validate reduced cost and successful artifact publishing.
AI-written summary · about
chapters (11)
- 0:00Fine-tuning LLMs with ML orchestration
- 2:29Why fine-tune large language models
- 6:05Prompt engineering vs fine-tuning
- 9:59Rationale for fine-tuning LLMs
- 10:39Fine-tuning methods
- 12:26Training methods for LLMs
- 13:51Optimization techniques
- 17:42Cost-effective fine-tuning
- 18:27How ML orchestrators work
- 21:16Demo: Fine-tuning with Flyte
- 23:41Summary
related talks
33mFine-Tuning Large Language Models with Declarative ML Orchestration
16mStreamline LLM Fine-tuning on Kubernetes With Kubeflow LLM Trainer
31mDeploying Large Language Models with Ease
39mThe Future is Fine-Tuned: Deploying Task-specific LLMs
32mRunning and Fine-tuning Open Source LLMs on Google Kubernetes Engine
21mStreamline LLM Fine-Tuning on Kubernetes with Kubeflow LLM Trainer
30mLearning to wield large language models
53mLLM Quality Optimization Bootcamp